Leveraging earth observational data products and machine learning to enhance urban building energy modeling (UBEM) with microclimate effects

Urban Building Energy Modeling (UBEM) is a powerful tool used for sustainable design, urban planning, and efficient energy management, as it provides essential insights into the building energy consumption patterns. However, current UBEM methodologies often lack urban-specific microclimate data, leading to discrepancies between modeled and actual energy consumption. This research develops a bottom-up statistical UBEM framework that combines and integrates earth observational climate data, climate reanalysis products, and annual energy usage data, measured by the Seattle Energy Benchmarking Dataset, to capture the impacts of microclimates on urban building energy performance. Using machine learning techniques and Seattle, Washington, USA as a proof of concept, our results demonstrate that incorporating urban-specific microclimate data substantially enhances building energy modeling prediction accuracy. Specifically, three model variable schemas are compared; the optimal model incorporating earth observational data achieved a 0.16 (from 0.55 to 0.71) increase in testing R2 over the model that did not include any climate data inputs, and a 0.056 (from 0.66 to 0.71) increase in testing R2, over the model that included TMY3 climate data inputs. These findings validate the use of earth observational datasets for urban building energy modeling to include microclimate effects. Furthermore, machine learning algorithms outperform traditional linear methods, with respective ordered rankings: CATBoost, XGBoost, Random Forest, Decision Trees, and Linear Regression. Our study underscores the importance of integrating microclimate data into UBEM frameworks and advocates for the expanded use of earth observational and remote sensing datasets for mitigation of simulation-to-reality discrepancies; to ultimately inform more accurate energy-driven design and planning strategies at the city level.